揭示智能体大模型安全漏洞，提出四层分类体系与七大未解难题。
On Understanding, Identifying, and Mitigating Vulnerabilities in Agentic Large Language Models
- 构建感知、思维、动作、交互四层漏洞分类框架
- 发现攻击研究是防御的3.9倍，代码执行安全仅占3.5%
- 指出架构耦合导致漏洞跨层传播，适合安全研究者参考
大语言模型已从无状态对话接口演变为具备多步规划、工具调用、代码执行和持久记忆的自主智能体。当这些智能体拥有真实权限——调用API、修改文件、查询数据库时，单个推理步骤被攻破就可能引发未经授权的数据访问、不可逆状态变更或级联故障。然而，安全研究尚未跟上发展步伐。我们基于PRISMA 2020指南，在六个数据库中系统综述743篇文献，筛选出85篇（2023–2025）关于智能体大模型安全的研究。攻击研究数量是防御工作的3.9倍；感知层漏洞（提示注入、越狱、对抗扰动）占66%，而行动层漏洞（工具误用、代码注入、沙箱逃逸）仅占4.7%，与真实风险不匹配。代码执行安全仅占3.5%，工具增强型智能体相关研究占12%。本文提出一个四层分类体系，涵盖感知、大脑、行动、交互四个层级，映射13类漏洞，并识别出七个围绕隔离机制的开放问题。智能体大模型的安全缺陷源于架构耦合，弱隔离使得漏洞可跨层传播。
原文摘要 · Abstract (English)
Large Language Models (LLMs) have undergone a shift from stateless conversational interfaces to autonomous agents capable of multi-step planning, tool invocation, code execution, and maintaining persistent memory. When these agents operate with real-world privileges---calling APIs, modifying files, and querying databases---a compromised reasoning step can trigger unauthorized data access, irreversible state changes, or cascading failures, yet the security research community has not kept pace. To quantify the state of the field, we conducted a systematic literature review under PRISMA 2020 guidelines across six databases, screening 743 records and retaining 85 papers (2023--2025) on agentic LLM security. Attack research outpaces defense work by 3.9:1. Perception-layer vulnerabilities (prompt injection, jailbreaking, adversarial perturbations) dominate, accounting for 66\% of papers, while action-layer vulnerabilities (tool misuse, code injection, sandbox escape) appear in only 4.7\%, misaligned with real-world risk. Code execution security accounts for 3.5\%, and tool-augmented agents 12\%. We contribute a four-layer taxonomy mapping 13 vulnerability types across perception, brain, action, and interaction layers, and identify seven open problems centered on containment. Agentic LLM insecurity stems from architectural coupling, where weak isolation allows vulnerabilities to propagate across layers.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。